The final publication is available in: Accident Analysis and Prevention. 2019, 123 (February), 1-11.
10.1016/j.aap.2018.11.002
Vehicle registration year, age, and weight – Untangling the effects on crash risk
Alena Høye
Institute of Transport Economics, Gaustadalléen 21, 0349 Oslo, Norway
The aim of the present study was to investigate the effects of passenger cars’ first year of registration, weight, and age on the number of killed or seriously injured (KSI) car drivers, pedestrians, and cyclists. Poisson regression models were developed of injury crashes involving passenger cars in Norway in 2000-2016, with the following predictor variables: The cars’ first year of registration and weight, either crash year or car age, the drivers age and gender, and in models for car-car collisions the crash partner cars’ weight and either registration year or age.
The results show that there are fewer KSI car drivers in more recent, newer, and heavier cars. It is estimated that the number of KSI car drivers in all types of crashes on average decreases by 6.7% for each consecutive registration year (-7.2% in car-car collisions and -6.0% in single vehicle crashes), increases by 3.7% for each consecutive year of age (+2.1% in car-car collisions and +5.3% in single vehicle crashes), and decreases by 4.9% on average for each 100 kg weight increase (-11.1% in car-car collisions and -2.3% in single vehicle crashes). In car-car collisions there are fewer KSI car drivers when the crash partner car is more recent (-4.4%
for each consecutive registration year), and more KSI car drivers when the crash partner car is older (+4.1% for each consecutive year of age), or heavier (+6.8% per 100 kg weight increase). In collisions with pedestrians or cyclists, there are fewer KSI pedestrians/cyclists when the car is more recent (-3.3% per consecutive registration year) and more KSI pedestrians/cyclists when the car is heavier (+4.6% per 100 kg weight increase). Due to the large effects of safety improvements in more recent cars, an increased renewal rate in the passenger car fleet can be expected to contribute to large safety improvements. The increasing weight of more recent cars may contribute to improved safety for those who drive heavier cars, but overall the effect of increasing weight is probably small or even negative because heavier vehicles impose greater risk on other car drivers, pedestrians, and cyclists.
Keywords: Registration year; car age; vehicle weight; fatality, serious injury
1. Introduction
Passenger cars have become considerably safer over the past decades, partly because of improvements of crashworthiness (Anderson & Searson, 2015), partly because of active safety systems (Glassbrenner, 2012), and partly because of increasing weight (Broughton, 2008).
Additionally, road safety in general has improved over time, which is partly due to safer cars, but also to other factors such as improved infrastructure, reduced drunk driving (Elvik et al., 2012), improved medical services (Schoell, 2014), increased seat belt use (Høye, 2016), and reduced speed (Sagberg & Bjørnskau, 2016).
There are relatively few other studies that have investigated the relationships between passenger cars’ registration year, car age, weight and crash or injury risk that have controlled for relevant confounding variables, and still fewer studies of the effects of the properties of crash partner vehicles on the driver of the own car or of the effects of the own cars’
registration year, age and weight on injury or crash risk among pedestrians and cyclists.
A general problem when investigating the effects of registration year, car age, and weight on crash or injury risk is that there are strong relationships between these variables and that all variables are strongly correlated to crash year. In later crash years, there are more cars from later registration years, and cars from later registration years are on average both newer and heavier. When the effects of registration year or car age are investigated without statistical control for crash year and either car age or registration year, the results for registration year or car age may be partly due to general safety improvement over time and the relationship between registration year and car age (e.g. Broughton, 2012; Martin & Lenguerrand, 2008;
Newstead et al., 2013). When the effects of vehicle weight on crash or injury risk are investigated without control for registration year (e.g. Anderson & Auffhammer, 2014;
Adolph et al., 2015) the results may be affected by the relationship between weight and model year (heavier cars are on average from later years). There are also relationships between properties of the cars and the drivers’ age, gender and behavior. For example, older cars are
more often driven by young males, and by drivers who are drunk, speeding, and/or not wearing seat belts (Høye, 2017A; White, 2003).
Another general problem when investigating effects of registration year, car age, and weight is that the only available exposure date often is crash involvement. Only very few studies of have used vehicle kilometers as exposure variable. Broughton (2008, 2012) and Farmer (2005) have used numbers of registered vehicles as exposure variable. Since older cars on average drive less than newer cars, such studies will overestimate the exposure of older cars and consequently underestimate the crash or injury risk of older cars. Wenzel (2013) and Pucket & Kindelberger (2016) have used an induced exposure methodology as an approximation for vehicle
kilometers. Most other studies of the effects of registration year, age or weight have therefore investigated effects on the number of fatally or seriously injured drivers per crash.
The aim of the present paper is to investigate the effect of passenger cars’ year of first registration, age, and weight on the number of KSI car drivers and pedestrians/cyclists in different types of crashes, based on Norwegian crash data from 2000-2016, with statistical control for millions of vehicle kilometers, crash year and the drivers’ age and gender. Crash types included in the study are all crashes, single vehicle crashes, car-car collisions, and collisions with pedestrians or cyclists. In car-car collisions, the effects of the other cars’
registration year or age, and weight were investigated as well. Dependent variables are the number of killed or seriously injured (KSI) car drivers (all crashes, single vehicle crashes and car-car collisions), and the number of KSI pedestrians or cyclists (collisions with
pedestrians/cyclists). Since the effects of registration year, car age, and crash year cannot be investigated simultaneously, models were developed with registration year and either crash year or car age among the predictor variables, and relationships between these variables were taken into account.
2. Method
Poisson regression models were developed for all crashes, single vehicle crashes, car-car collisions, and collisions with pedestrians or cyclists. In Poisson regression the logarithm of the expected values of a dependent variable (numbers of KSI in the present study) is modeled as a linear combination of predictor variables according to the following formula:
log (𝐸𝐸(𝑌𝑌|𝑥𝑥)) = 𝛼𝛼+� 𝛽𝛽𝑖𝑖 ∗ 𝑥𝑥𝑖𝑖 𝑖𝑖
E(Y│x) is the expected number of KSI, α is an intercept, xi are the predictor variables, and βi
the coefficients for the predictor variables. In the present study, all predictor variables are defined as sets of dummy variables. The estimated coefficient for a dummy predictor variable can be interpreted as the natural logarithm of the relative number of the dependent variable when the predictor is one (present), compared to zero (absent). For example, a coefficient of 0.5 means that the expected value of the dependent variable is 65% higher when the predictor variable is present than when it is absent (Ln(1.65) = 0.5 or e0.5 = 1.65).
The dependent variable in the models for the first three crash types is the number of KSI car drivers. The study focuses on passenger car drivers because each car normally has one driver, which makes it possible to apply exposure data for cars to car drivers. Exposure data is available at the required level of detail for cars, but not for passengers. In the models for collisions with pedestrians/cyclists the number of KSI pedestrians/cyclists is the dependent variable.
Predictor variables in all models are the cars’ year of first registration year, the cars’ weight, the drivers’ age and gender, and exposure (millions of vehicle kilometers). Additionally, either crash year or the cars’ age are predictor variables. Both crash year and car age cannot be predictor variables because among the three variables registration year, car age, and crash year, one is always a function of the other two. In car-car collisions, the crash partners’ (the other cars’) registration year (car age in the models with the own cars’ age among the predictor
as an exposure variable, i.e. the coefficient for the natural logarithm of million vehicle kilometers is fixed at one such that the assumed relationship between the total number of vehicle kilometers and KSI is linear. All models were calculated in Stata (version 14.2).
The unit of analysis is a group of cars, with each group containing all Norwegian cars from registration year A (10 groups), in weight class B (four weight classes), with a driver in group C (eight groups according to age and gender), in crash year D (years 2000-2016), with a crash partner from registration year E in weight class F (crash partner registration year and weight only in the model for car-car collisions). In the models with car age among the predictor variables, crash year is replaced by car age, calculated as the difference between crash year and the average registration year in each group of registration year.
For each unit of analysis and each type of crash, the numbers of KSI car drivers
(pedestrians/cyclists in collisions with pedestrians/cyclists) have been retrieved from official injury crash statistics published by Statistics Norway (www.ssb.no). The total number of vehicle kilometers has been estimated based on a cohort model of the national car fleet (Fridstrøm et al., 2016). The model describes the annual mileages for cars of different ages, weight classes, and engine types, based on odometer readings from the registry of periodic vehicle inspection. Based on these data relative annual mileages were calculated for cars in each group of car age (yearly) and weight, and relative annual mileages in each group of registration year and weight were calculated for each year in the study period. The distribution of annual mileages on car age and weight categories is assumed to remain unchanged over time.
Relative annual driving lengths for drivers in each group of age and gender were retrieved from Bjørnskau (2015) who has analyzed data from national travel surveys among
representative samples of the Norwegian population. Data are available every fourth year. The distribution of relative annual driving lengths on the groups by age and gender are relatively unchanged over time and are in the present study assumed to be the same in all years.
Total annual driving lengths with passenger cars in Norway for each year during the study period were retrieved from Farstad (2016). The total annual driving lengths were then
distributed over all combinations of the cars registration year and age and the drivers age and gender, assuming no interaction effect between car and driver characteristics. The latter assumption is not likely to be true. For example, young drivers are on average driving older cars than older drivers. However, no information was available to link relative driving lengths between driver of different ages and gender and cars of different registration year and weight.
3. Results
Table 1 shows the numbers of KSI car drivers in all crashes, single vehicle crashes, and car-car collisions and the numbers of KSI pedestrians and cyclists in collisions between a car and a pedestrian or cyclist for all predictor variables. An overview of the developed models is shown in Table 2. As a measure of model fit McFadden's log-likelihood ratio index Pseudo R-squared has been calculated in Stata. Pseudo R-squared approaches one as the fit of the model
becomes better and zero as the fit becomes weaker. Table 3 and Table 4 show the results from Poisson regression models for all crash types, in Table 3 with crash year (and not the age of the own car) among the predictor variables, and in Table 4 with the own cars’ age (and not crash year) among the predictor variables. The models for each of the dependent variables contain all variables in the tables as predictor variables, i.e. the results for each predictor variable describe the relationship between predictor and dependent variable when all other variables in the model are held constant (i.e. statistically controlled for). The coefficients for the own and the other cars’ weight and for the drivers’ age and gender are very similar between the two types of models.
Table 1: Descriptive statistics; millions of vehicle kilometers, KSI car drivers (own car) in all crashes, single vehicle crashes, and car-car collisions, and KSI pedestrians/cyclists in collisions with cars (KSI/million vehicle kilometers above median in bold).
All crashes SV crashes Car-car collisions Ped./cyc. collisions Mill. veh.
km KSI car
drivers KSI/mill.
veh. km KSI car
drivers KSI/mill.
veh. km KSI car
drivers KSI/mill.
veh. km KSI
ped./cyc. KSI/mill.
veh. km
Registration year (own car)
X-1979 3,264 52 0.0159 21 0.0064 14 0.0043 10 0.0031
1980-1990 45,941 1,730 0.0377 729 0.0159 463 0.0101 363 0.0079
1991-1995 56,627 1,252 0.0221 509 0.0090 300 0.0053 301 0.0053
1996-2000 137,216 1,545 0.0113 601 0.0044 369 0.0027 516 0.0038
2001-2003 77,818 532 0.0068 208 0.0027 119 0.0015 233 0.0030
2004-2006 78,716 385 0.0049 133 0.0017 83 0.0011 162 0.0021
2007-2009 59,507 209 0.0035 75 0.0013 50 0.0008 111 0.0019
2010-2011 31,770 103 0.0032 32 0.0010 27 0.0008 74 0.0023
2012-2013 22,860 68 0.0030 22 0.0010 19 0.0008 64 0.0028
2014-2016 13,511 31 0.0023 13 0.0010 9 0.0007 42 0.0031
Weight (own car)
0-1199 kg 151,070 2,510 0.0166 939 0.0062 704 0.0047 590 0.0039
1200-1399 kg 159,732 1,936 0.0121 781 0.0049 459 0.0029 578 0.0036
1400-1599 kg 128,357 970 0.0076 410 0.0032 204 0.0016 391 0.0030
1600+ kg 88,072 491 0.0056 213 0.0024 86 0.0010 317 0.0036
Registration year (other car)
X-1979 10 0.0031
1980-1990 273 0.0059
1991-1995 271 0.0048
1996-2000 428 0.0031
2001-2003 150 0.0019
2004-2006 154 0.0020
2007-2009 88 0.0015
2010-2011 34 0.0011
2012-2013 32 0.0014
2014-2016 13 0.0010
Weight (other car)
0-1199 kg 376 0.0025
1200-1399 kg 465 0.0029
1400-1599 kg 340 0.0026
1600+ kg 272 0.0031
Driver
Female, 18-24 years 16,425 381 0.0232 156 0.0095 105 0.0064 78 0.0047 Female, 25-44 years 84,020 623 0.0074 181 0.0022 172 0.0020 217 0.0026 Female, 45-64 years 63,618 504 0.0079 137 0.0022 135 0.0021 154 0.0024
Female, 65+ years 14,780 247 0.0167 83 0.0056 73 0.0049 90 0.0061
Male, 18-24 years 26,960 1,243 0.0461 727 0.0270 233 0.0086 285 0.0106 Male, 25-44 years 147,877 1,460 0.0099 615 0.0042 340 0.0023 473 0.0032 Male, 45-64 years 133,956 786 0.0059 250 0.0019 214 0.0016 321 0.0024 Male, 65+ years 39,594 663 0.0167 194 0.0049 181 0.0046 258 0.0065
Crash year
2000 26,136 550 0.0210 212 0.0081 154 0.0059 151 0.0058
2001 26,772 437 0.0163 178 0.0066 111 0.0041 135 0.0050
2002 27,539 497 0.0180 194 0.0070 126 0.0046 136 0.0049
2003 28,022 437 0.0156 179 0.0064 107 0.0038 134 0.0048
2004 28,570 408 0.0143 145 0.0051 107 0.0037 120 0.0042
2005 29,300 385 0.0131 153 0.0052 91 0.0031 113 0.0039
2006 29,799 374 0.0126 155 0.0052 83 0.0028 126 0.0042
2007 30,936 346 0.0112 138 0.0045 85 0.0027 92 0.0030
2008 31,296 358 0.0114 154 0.0049 86 0.0027 108 0.0035
2009 31,727 316 0.0100 116 0.0037 83 0.0026 80 0.0025
2010 32,120 326 0.0101 127 0.0040 68 0.0021 64 0.0020
2011 32,725 276 0.0084 108 0.0033 64 0.0020 86 0.0026
2012 33,115 266 0.0080 109 0.0033 61 0.0018 85 0.0026
2013 33,604 267 0.0079 94 0.0028 73 0.0022 109 0.0032
2014 34,434 230 0.0067 88 0.0026 58 0.0017 118 0.0034
2015 35,409 225 0.0064 103 0.0029 54 0.0015 111 0.0031
2016 35,727 209 0.0058 90 0.0025 42 0.0012 108 0.0030
Car age
0-5 years 197,993 885 0.0045 302 0.0015 222 0.0011 484 0.0024 6-10 years 155,659 1263 0.0081 479 0.0031 279 0.0018 488 0.0031
11-15 years 94,212 1284 0.0136 497 0.0053 313 0.0033 380 0.0040
16-20 years 59,266 1810 0.0305 777 0.0131 477 0.0080 404 0.0068
21-25 years 12,833 520 0.0405 224 0.0175 127 0.0099 95 0.0074
26-30 years 3,300 81 0.0245 38 0.0115 18 0.0055 12 0.0036
31-35 years 2,804 55 0.0196 22 0.0078 15 0.0053 12 0.0043
36+ years 1,163 9 0.0077 4 0.0034 2 0.0017 1 0.0009
Total 527,213 5,907 0.0112 2,343 0.0044 1,453 0.0028 1,876 0.0036
Table 2: Model characteristics.
Among the
predictors N* Pseudo
R-squared
All crashes Crash year 3,680 0.4073
Car age 3,680 0.4064
Single vehicle crashes Crash year 3,680 0.3923
Car age 3,680 0.3910
Car-car collisions Crash year 104,320 0.1337 Car age 104,320 0.1328 Collisions with ped./cycl. Crash year 3,680 0.1365
Car age 3,680 0.1312
*The number of rows in the data file, depending on the grouping of the data according to the predictor variables.
Table 3: Model results, coefficients from Poisson regression models with crash year (and not car age) among the predictor variables for all crashes, single vehicle crashes, and car-car collisions (statistically significant coefficients in bold).
All crashes Single vehicle crashes Car-car collisions Ped./cyc. collisions
Coeff. p Coeff. p Coeff. p Coeff. p
Registration year (own car)
X-1979 2.285 .000 2.560 0.000 1.919 0.000 0.365 0.315
1980-1990 3.122 .000 3.439 0.000 2.749 0.000 1.312 0.000
1991-1995 2.551 .000 2.796 0.000 2.106 0.000 0.906 0.000
1996-2000 1.823 .000 1.992 0.000 1.404 0.000 0.520 0.004
2001-2003 1.269 .000 1.391 0.000 0.843 0.019 0.242 0.183
2004-2006 0.888 .000 0.836 0.005 0.473 0.192 -0.181 0.326
2007-2009 0.509 .010 0.470 0.124 0.222 0.549 -0.313 0.099
2010-2011 0.393 .059 0.188 0.573 0.200 0.611 -0.182 0.363
2012-2013 0.280 .200 0.093 0.793 0.143 0.727 -0.099 0.624
2014-2016 (ref.) (ref.) (ref.)
Weight (own car)
0-1199 kg 0.418 .000 0.206 0.008 0.899 0.000 -0.349 0.000
1200-1399 kg 0.499 .000 0.401 0.000 0.817 0.000 -0.160 0.024
1400-1599 kg 0.421 .000 0.418 0.000 0.604 0.000 -0.100 0.188
1600+ kg (ref.) (ref.) (ref.) (ref.)
Registration year (other car)
X-1979 0.703 0.108
1980-1990 1.365 0.000
1991-1995 1.213 0.000
1996-2000 0.814 0.007
2001-2003 0.403 0.189
2004-2006 0.479 0.115
2007-2009 0.230 0.458
2010-2011 -0.050 0.883
2012-2013 0.236 0.482
2014-2016 (ref.)
Weight (other car)
0-1199 kg -0.510 0.000
1200-1399 kg -0.201 0.009
1400-1599 kg -0.126 0.123
1600+ kg (ref.)
Driver
Female, 18-24 years 0.220 .001 0.567 0.000 0.214 0.081 -0.373 0.004 Female, 25-44 years -0.902 .000 -0.902 0.000 -0.901 0.000 -0.978 0.000 Female, 45-64 years -0.769 .000 -0.842 0.000 -0.787 0.000 -1.001 0.000 Female, 65+ years 0.113 .131 0.239 0.069 0.207 0.137 -0.020 0.872 Male, 18-24 years 0.926 .000 1.626 0.000 0.538 0.000 0.429 0.000 Male, 25-44 years -0.628 .000 -0.255 0.002 -0.799 0.000 -0.766 0.000 Male, 45-64 years -1.099 .000 -1.012 0.000 -1.106 0.000 -1.020 0.000
Male, 65+ years (ref.) (ref.) (ref.) (ref.)
Crash year
2000 -0.476 .000 -0.807 0.000 -0.797 0.000 -0.129 0.366
2001 -0.569 .000 -0.825 0.000 -0.902 0.000 -0.161 0.265
2002 -0.429 .000 -0.728 0.000 -0.763 0.000 -0.174 0.226
2003 -0.468 .000 -0.708 0.000 -0.787 0.000 -0.148 0.301
2004 -0.435 .000 -0.804 0.000 -0.639 0.002 -0.207 0.154
2005 -0.387 .000 -0.628 0.000 -0.651 0.002 -0.211 0.150
2006 -0.302 .001 -0.484 0.000 -0.587 0.005 -0.040 0.779
2007 -0.320 .000 -0.541 0.000 -0.473 0.023 -0.355 0.020
2008 -0.155 .087 -0.278 0.043 -0.261 0.206 -0.109 0.453
2009 -0.192 .038 -0.463 0.001 -0.167 0.417 -0.369 0.018
2010 -0.090 .324 -0.290 0.041 -0.256 0.226 -0.560 0.001
2011 -0.167 .075 -0.345 0.018 -0.174 0.412 -0.244 0.107
2012 -0.110 .242 -0.223 0.125 -0.088 0.679 -0.234 0.120
2013 -0.012 .902 -0.257 0.086 0.214 0.293 0.029 0.835
2014 -0.079 .414 -0.225 0.137 0.093 0.656 0.104 0.444
2015 -0.023 .808 0.023 0.872 0.128 0.535 0.027 0.843
2016 (ref.) (ref.) (ref.) (ref.)
Mill. vehicle km 1.000 1.000 1.000 1.000
Constant -5.959 .000 -7.051 .000 -7.585 .000 -5.090 .000
Table 4: Model results, coefficients from Poisson regression models with car age (and not crash year) among the predictor variables for all crashes, single vehicle crashes, and car-car collisions (statistically significant coefficients in bold).
All crashes Single vehicle crashes Car-car collisions Ped./cyc. collisions
Coeff. p Coeff. p Coeff. p Coeff. p
Registration year (own car)
X-1979 0.702 0.005 0.189 0.631 1.103 0.024 -0.044 0.913
1980-1990 2.078 0.000 1.842 0.000 2.235 0.000 1.006 0.000
1991-1995 1.795 0.000 1.611 0.000 1.752 0.000 0.623 0.001
1996-2000 1.249 0.000 1.068 0.000 1.158 0.001 0.274 0.106
2001-2003 0.837 0.000 0.670 0.020 0.674 0.054 0.018 0.919
2004-2006 0.569 0.002 0.281 0.335 0.374 0.289 -0.386 0.028
2007-2009 0.301 0.119 0.068 0.821 0.197 0.588 -0.499 0.006
2010-2011 0.275 0.179 -0.067 0.839 0.223 0.563 -0.277 0.152
2012-2013 0.227 0.295 -0.054 0.877 0.214 0.596 -0.099 0.618
2014-2016 (ref.) (ref.) (ref.) (ref.)
Weight (own car)
0-1199 kg 0.418 0.000 0.206 0.008 0.899 0.000 -0.345 0.000
1200-1399 kg 0.500 0.000 0.401 0.000 0.817 0.000 -0.161 0.023 1400-1599 kg 0.420 0.000 0.418 0.000 0.603 0.000 -0.101 0.182
1600+ kg (ref.) (ref.) (ref.) (ref.)
Weight (other car)
0-1199 kg -0.512 0.000
1200-1399 kg -0.203 0.008
1400-1599 kg -0.126 0.122
1600+ kg (ref.)
Driver
Female, 18-24 years 0.222 0.001 0.567 0.000 0.218 0.077 -0.380 0.003 Female, 25-44 years -0.902 0.000 -0.902 0.000 -0.902 0.000 -0.979 0.000 Female, 45-64 years -0.769 0.000 -0.842 0.000 -0.790 0.000 -1.002 0.000 Female, 65+ years 0.110 0.142 0.239 0.069 0.204 0.142 -0.006 0.964 Male, 18-24 years 0.926 0.000 1.626 0.000 0.538 0.000 0.429 0.000 Male, 25-44 years -0.627 0.000 -0.256 0.002 -0.799 0.000 -0.772 0.000 Male, 45-64 years -1.098 0.000 -1.012 0.000 -1.106 0.000 -1.028 0.000
Male, 65+ years (ref.) (ref.) (ref.) (ref.)
Car age (own car) 0.036 0.000 0.052 0.000 0.020 0.003 0.006 0.289
Car age (other car) 0.040 0.000
Mill. vehicle km 1.000 1.000 1.000 1.000
Constant -5.991 0.000 -7.089 0.000 -7.530 0.000 -5.070 0.000
3.1 Registration year
The results in Table 1 and the model results (Table 3 and Table 4) show consistently for all crash types that there are fewer KSI car drivers in cars from later registration years than in cars from earlier years, and fewer KSI pedestrians/cyclists in collisions with cars from later
registration years. However, in cars registered before 1980 there are fewer KSI drivers than in cars from 1980-1990 which may be a veteran car effect. Many drivers of very old cars are careful with their cars, drive only at low speed under low-risk conditions, and possibly less than assumed in the models.
Figure 1 shows the relationships between registration year and relative numbers of KSI, both unadjusted (per million vehicle kilometers as shown in Table 1) and adjusted (based on model results with statistical control for all other variables in the models; Table 3 and Table 4). The estimated changes of numbers of KSI from registration year 1980-1990 to 2014-2016 and from registration year X to X+1 for all types of crashes, are shown in Table 5, based on the unadjusted results (Table 1) and the adjusted results with crash year (Table 3) or car age (Table 4) among the predictor variables.
Figure 1: Relationships between the own and other cars’ registration year and relative numbers of KSI car drivers (own car) or pedestrians/cyclists; unadjusted and adjusted results (rel. number of KSI is in each diagram set equal to one for cars from 1980-1990).
0.0 0.2 0.4 0.6 0.8 1.0
Own cars' registration year All crashes: KSI car drivers
Unadjusted Adjusted (crash year) Adjusted (car 0.0 age)
0.2 0.4 0.6 0.8 1.0
Own cars' registration year Single vehicle crashes: KSI car drivers
0.0 0.2 0.4 0.6 0.8 1.0
Own cars' registration year Car-car collisions: KSI car drivers
Unadjusted Adjusted (crash year) Adjusted (car 0.0 age)
0.2 0.4 0.6 0.8 1.0
Other cars' registration year Car-car collisions: KSI car drivers
0.0 0.2 0.4 0.6 0.8 1.0
Own cars' registration year Collisions with ped./cycl.: KSI ped./cycl.
Unadjusted Adjusted (crash year) Adjusted (car age)
Table 5: Estimated changes of numbers of KSI from registration year 1980-1990 to 2014-2016 and from registration year X to X+1 for all types of crashes, unadjusted and adjusted.
Crash type: All crashes Single vehicle
crashes Car-car collisions Car-car collisions Collisions with pedestrians/cyclists
Effect of reg. year: Own car Own car Own car Other car Own car
Effect on KSI: Drivers (own car) Drivers (own car) Drivers (own car) Drivers (own car) Pedestrians/ cyclists From registration year 1980-1990 to 2014-2016
Unadjusted -94 % -94 % -93 % -83 % -61 %
Adjusted (crash year pred.) -96 % -97 % -94 % -74 % -73 %
Adjusted (car age pred.) -87 % -84 % -89 % -63 %
From registration year X to X+1
Unadjusted -8.9 % -8.8 % -8.5 % -5.7 % -3.1 %
Adjusted (crash year pred.) -9.9 % -10.8 % -8.8 % -4.4 % -4.3 %
Adjusted (car age pred.) -6.7 % -6.0 % -7.2 % -3.3 %
According to the unadjusted results and the adjusted results with crash year among the
predictor variables, Figure 1 and Table 5 show that the numbers of KSI car drivers per million vehicle kilometers have decreased by over 90% in cars from 2014-2016 compared to cars from 1980-1990 in all crashes, single vehicle crashes and car-car collisions. The slope of the
decrease has changed less than it may appear in Figure 1 because more years are grouped together during earlier years than during later years. However, even when taking into account the different sizes of the age groups, the decrease seems to have leveled off during the later years.
The adjusted effects of the own cars’ registration year with crash year among the predictor variables are slightly larger than the unadjusted effects. This is contrary to expectation because both weight and crash year are statistically controlled for. The average weight of new cars has increased from about 1100 kg in 1970-1990 to about 1460 in 2015 and 2016. Increasing weight was found to be related to decreasing injury risk for the driver, both in the present and in other studies (section 3.3). Crash year is also related to crash and injury risk, the number of KSI per million vehicle kilometers has decreased by 72% from 2000 to 2016 (Table 1).
Possible explanations for the large adjusted effects of registration year with crash year among the predictor variables are discussed in section 4.1.
The adjusted effects of the own cars’ registration year with car age (and not crash year) among the predictor variables in Figure 1 are still large, but consistently smaller than the unadjusted
effects and the adjusted effects with crash year among the predictor variables, except in collisions with pedestrians/cyclists. In all crashes they indicate that the number of KSI car drivers in the own car decreases on average by 6.7% for each consecutive registration year.
In car-car collisions, more recent own and other cars are related to fewer KSI car drivers (in the own car), with the effect of the own cars’ registration year being larger than the other cars’
registration year. Whether the effect of the own cars’ registration year on KSI car drivers is larger or smaller in car-car collisions or in single vehicle crashes, differs between the models with crash year and car age among the predictor variables.
In collisions with pedestrians/cyclists, the number of KSI pedestrians/cyclists was found to increase for cars of the latest registration years, compared to cars from 2004-2009 in the models with car age among the predictor variables, while the differences between the latest registration years in the models with crash year among the predictor variables is not
statistically significant.
3.2 Car age
The relationships between car age and relative numbers of KSI car drivers and
pedestrians/cyclists are shown in Figure 2, both unadjusted (as in Table 1) and adjusted (as in Table 4), disaggregated by year. The relative number of KSI is in each diagram set equal to one for one-year-old cars.
Figure 2: Relationships between the own and other cars’ age and relative numbers of KSI car drivers (own car) or pedestrians/cyclists; unadjusted and adjusted results (rel. number of KSI is in each diagram set equal to one for one-year-old cars).
The results in Table 1 and Figure 2 show that there are more KSI car drivers in older cars up to an age of about 25 years. In cars above 25 years, the number of KSI decreases with increasing age. This may be the same veteran car effect as for cars from the earliest registration years.
The adjusted results show far weaker increases of the number of KSI with increasing car age than the unadjusted effects. A large part of the unadjusted effects of age on KSI is likely to be due to the effect of registration year which is strongly related to both KSI and car age. Cars from 2014-2016 are on average 1.4 years old, while cars from 1980-1990 on average are 19.6 years old.
The strongest increase of KSI with increasing car age was found in single vehicle crashes. The adjusted effects may be somewhat underestimated because of the decrease of KSI in the
0 5 10 15 20 25 30
0 5 10 15 20 25 30 35
Car age (own car) All crashes: KSI car drivers
Unadjusted Adjusted 0 5 10 15 20 25 30
0 5 10 15 20 25 30 35
Car age (own car) Single vehicle crashes: KSI car drivers
Unadjusted Adjusted
0 5 10 15 20 25 30
0 5 10 15 20 25 30 35
Car age (own car) Car-car collisions: KSI car drivers
Unadjusted Adjusted 0 5 10 15 20 25 30
0 5 10 15 20 25 30 35
Car age (other car) Car-car collisions: KSI car drivers
0 5 10 15 20 25 30
0 5 10 15 20 25 30 35 Car age (own car) Collisions with ped./cycl.: KSI ped./cycl.
Unadjusted Adjusted +1 year: +3.7 % KSI (adj.) +1 year: +5.3 %
KSI (adj.)
+1 year: +2.1 % KSI (adj.) +1 year: +4.1 % KSI (adj.)
+1 year: +0.6 % KSI ped./cycl. (adj.)
oldest cars. However, only 2.5% of all KSI drivers had driven cars above 25 years of age, while 39.4% had a car that was 16-25 years old. The adjusted effects are therefore not likely to be strongly affected by the decreasing numbers of KSI in the oldest cars.
The age of the crash partner in car-car collisions has a far weaker effect on KSI car drivers than the age of the own car according to the unadjusted results. However, the adjusted effect of the other cars’ age is about twice as large as the adjusted effect of the own cars’ age (+4.1%
vs. +2.1% KSI per additional year).
In collisions with pedestrians/cyclists the effect of the own cars’ age on KSI
pedestrians/cyclists is far smaller than the effects on KSI drivers in other crashes and not statistically significant.
3.3 Weight
The relationships between the cars’ weight and relative numbers of KSI car drivers and pedestrians/cyclists are shown in Figure 3, both unadjusted (based on the results in Table 1) and adjusted (based on the results in Table 3). The coefficients for weight are almost identical to the third decimal in the models with crash year and car age among the predictors. The relative number of KSI is in all diagrams set equal to one for the lightest cars. Additionally, exponential trend functions of the adjusted effects are shown in each diagram as well as the average effects of a weight increase by 100 kg, based on the trend functions.
Figure 3: Relationships between the own and other cars’ weight and relative numbers of KSI car drivers (own car) or pedestrians/cyclists; unadjusted and adjusted results (rel. number of KSI is in each diagram set equal to one for the lightest cars).
The adjusted effects of the own cars’ weight on KSI car drivers are consistently smaller than the corresponding unadjusted effects. This is probably mainly due to the statistical control for registration year because more recent cars are on average both heavier and safer than cars from earlier years.
In all crashes and in and single vehicle crashes non-linear relationships were found between weight and KSI car drivers. The trend functions and the estimated average changes of the number of KSI per 100 kg weight increase are therefore somewhat misleading for these crashes. Regardless of this, the adjusted results indicate that the effect of the own cars’ weight
Unadjusted Adjusted (crash year) Trend
Unadjusted Adjusted (crash year) Exp. trend (adj., crash year) 0 %
-27 %
-54 %
-66 %
0 % 8 %
0.3 %
-34 %
-100 % -80 % -60 % -40 % -20 % 0 % 20 % 40 %
0-1199 kg (ref.) 1200-
1399 kg 1400-
1599 kg 1600+ kg Own cars' weight
All crashes: KSI car drivers
Unadjusted Adjusted (crash year) Trend
0 %
-21 %
-48 %
-61 % 0 %
22 % 24 %
-19 %
-100 % -80 % -60 % -40 % -20 % 0 % 20 % 40 %
0-1199 kg (ref.) 1200-
1399 kg 1400-
1599 kg 1600+ kg Own cars' weight
Single vehicle crashes: KSI car drivers
0 %
-38 %
-66 %
-79 %
0 % -8 %
-26 %
-59 %
-100 % -80 % -60 % -40 % -20 % 0 % 20 % 40 %
0-1199 kg (ref.) 1200-
1399 kg 1400-
1599 kg 1600+ kg Own cars' weight
Car-car collisions: KSI car drivers
Unadjusted Adjusted (crash year) Trend0 %
16 % 4 %
0 % 24 %
36 % 47 %
67 %
-40 % -20 % 0 % 20 % 40 % 60 % 80 % 100 %
0-1199 kg (ref.) 1200-
1399 kg 1400-
1599 kg 1600+ kg Other cars' weight
Car-car collisions: KSI car driver
0 % -8 %
-23 % -8 %
0 %
21 % 28 % 42 %
-100 % -80 % -60 % -40 % -20 % 0 % 20 % 40 % 60 %
0-1199 kg (ref.) 1200-
1399 kg 1400-
1599 kg 1600+ kg Own cars' weight
Collisions with ped./cycl.: KSI ped./cycl.
+100 kg: -4.9 % KSI (adj. trend) +100 kg: -2.3 % KSI (adj. trend)
+100 kg: -11.1 % KSI (adj. trend) +100 kg: +6.8 % KSI (adj. trend)
+100 kg: +4.6 % KSI ped./cycl. (adj., trend)
on the number of KSI drivers of the own car is far larger in car-car collisions than in single vehicle crashes.
Increasing weight among crash partner vehicle is related to increasing numbers of KSI in the own car. The average absolute effect of a weight increase by 100 kg is somewhat smaller (+6.8%) than the corresponding effect of the own cars weight (-11.1%). In collisions with pedestrians and cyclists, heavier cars impose higher risk on pedestrians and cyclists than lighter cars according to the adjusted effects.
3.4 Driver
The results in Table 1, Table 3, and Table 4 show consistently that young drivers, especially young men, are more often KSI than other drivers, especially in single vehicle crashes. Older drivers are also more often KSI than drivers in the middle age groups, but less often than the youngest. While young men are more often KSI than young women, there are no large or statistically significant differences between men and women in the age groups above 25 years.
Young men impose also considerably larger risk to pedestrians and cyclists than other drivers.
The results from the present study are in accordance with results from other studies (e.g.
Bjørnskau, 2015; Kockelman & Kweon, 2002; Martin & Lenguerrand, 2008).
3.5 Crash year
In later crash years there are fewer KSI car drivers per million vehicle kilometers (Table 1), but the model results indicate that there are more KSI car drivers in later years (Table 3). A likely explanation is the statistical control for registration year. The model results show there are more KSI car drivers in later years if all else is equal, where all else includes the cars’ registration year. Cars from the same registration year are older in later crash years, and older cars can be assumed to have more KSI than newer cars, both according to the results from the present study, according to other studies (Farmer & Lund, 2015; Wenzel, 2013), and because of the relationship between car age and driver related risk factors (see section 4.2). Consequently, if all else, including the cars’ registration year, is equal, the higher car age in later years can be
expected to contribute to more KSI in later years. An example of the same type of effect is described by Kennedy (2005). In the example, an additional family room in residential houses contributes to decreasing house prices with statistical control for total square meters: Adding a family room while holding total square meters constant entails a reduction of either total square meters or number of rooms elsewhere in the house, which negatively affects the price.
Thus, the addition of the family room as such cannot be held responsible for the negative effect on house prices.
4. Discussion
The present study has investigated effects of passenger cars’ first year of registration, weight, and age on the number of KSI drivers of passenger cars and pedestrians/cyclists. A general problem in estimating the effects of registration year and car age is that both variables are highly correlated both to each other and to crash year. Moreover, two of these variables are always a function of the third and it is therefore not possible to develop models with all three variables as predictors. In the present study, controlling for crash year contributed to
unrealistically large effects of registration year, at the same time as the adjusted effects of crash year have the «wrong» sign. The most likely explanations are the large correlation between registration year and crash year (se section 4.1) and the fact that cars from the same
registration year are older in later crash years and that older cars have higher risk than newer cars (see section 3.5).
4.1 Registration year
Cars from later registration years were in the present study found to have far fewer KSI drivers than cars from earlier registration years (back to around 1980). The estimated
reduction of the number of KSI drivers in all crashes per year, i.e. from one registration year to the next, according to the model with crash year (and not car age) among the predictor variables (-9.9%) is likely to be overestimated because it is larger than the unadjusted effect (-
age (and not crash year) among the predictor variables, the estimated reduction is smaller (- 6.7%). Moreover, the decrease seems to have levelled off during later years.
A possible explanation for the large effect of registration year with statistical control for crash year is that the registration year predictor variable may «explain» a part of the effect of crash year. This is possible because registration year and crash year are highly correlated; cars from 2014-2016 crashed on average in 2015, while cars from 1980-1990 on average crashed in 2004.
Such an effect may also be a part of the explanation for the finding that the adjusted effects of crash year have the «wrong» sign (see section 3.5).
Directly comparable other studies were not found, i.e. studies that also have controlled for crash year or car age and that have used vehicle kilometers as exposure variable. The study that is most comparable to the present study (Broughton, 2012) found an average decrease of the number of fatally injured car drivers by 9.3% and an average decrease of seriously injured car drivers by 6.3% per registration year in car-car collisions from registration years 1988-1991 to 2004-2007 which is similar to the results from the present study (-8.8% or -7.2% KSI car drivers in car-car collisions, depending to the model). Broughton (2012) have investigated the effects on the number of fatally and seriously injured car drivers per registration year. Crash year (2001-2005) is not statistically controlled for.
Whether the effect of registration year is larger in single vehicle crashes or in car-car collisions is not quite clear in the results from the present study. The unadjusted effects of registration year are similar in single vehicle crashes and in car-car collisions, and the results from models with crash year and car age among the predictor variables yield contradicting results. Another study that has investigated effects of registration year in both types of crashes (Méndez et al., 2010) found no statistically significant effects in single vehicle crashes and an average decrease of the number of KSI car drivers per injured car driver by 2.9% per registration year
(registration years from pre 1985 to 2005 are included in the study) in car-car collisions. The effect is estimated based on an exponential trend function fitted to the results reported in the
